Alibaba Cloud Model Studio:Claude Code

Error: API Error: Unable to connect to API (ECONNRESET)

This error is caused by a network connection issue in the Claude Code client and is unrelated to configuration. It usually resolves on its own. Suggestions:

1. Wait a few minutes and try again.

2. Check that your network connection is working properly.

3. If you are using a proxy or VPN, disable it and try again.

4. Upgrade Claude Code to the latest version: npm install -g @anthropic-ai/claude-code@latest.

Error: Unable to connect to Anthropic services. Failed to connect to api.anthropic.com: ERR_BAD_REQUEST

This error indicates that Claude Code is trying to connect to the official Anthropic service rather than the Token Plan Team Edition service. This is usually because environment variables are not configured correctly or have not taken effect. Follow these steps to troubleshoot:

1. Check the configuration file: Verify that ~/.claude/settings.json has ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN configured correctly.

   # View the current configuration
   cat ~/.claude/settings.json

   Verify that the configuration looks like the following (replace with your actual API Key):

   {
       "env": {
           "ANTHROPIC_AUTH_TOKEN": "YOUR_API_KEY",
           "ANTHROPIC_BASE_URL": "https://token-plan.ap-southeast-1.maas.aliyuncs.com/apps/anthropic",
           "ANTHROPIC_MODEL": "qwen3.6-plus",
           "ANTHROPIC_SMALL_FAST_MODEL": "qwen3.6-plus",
           "ANTHROPIC_DEFAULT_HAIKU_MODEL": "qwen3.6-plus",
           "ANTHROPIC_DEFAULT_SONNET_MODEL": "qwen3.6-plus",
           "ANTHROPIC_DEFAULT_OPUS_MODEL": "qwen3.6-plus",
           "CLAUDE_CODE_SUBAGENT_MODEL": "qwen3.6-plus"
       }
   }

2. Check for environment variable conflicts: If ANTHROPIC_BASE_URL is set through both environment variables and the configuration file, ensure both point to the same Token Plan Team Edition address to avoid conflicts. Run the following command to check:

   echo $ANTHROPIC_BASE_URL

   If the output is empty or points to an address other than the Token Plan Team Edition endpoint, clear the environment variable or set it to the correct Token Plan Team Edition Base URL.

3. Verify hasCompletedOnboarding: Check that ~/.claude.json has hasCompletedOnboarding set to true. Otherwise, Claude Code will attempt to connect to the official Anthropic service for login verification on startup.

4. Open a new terminal: After modifying the configuration file, open a new terminal window and run the claude command for the configuration to take effect.

1. Context management

• Clean up regularly: Use /clear to periodically reset conversations, preventing old context from interfering with new tasks and saving tokens.

• Proactive compression: Use the /compact command to have Claude summarize key decisions and modified files, preserving core memory.

• Specify files explicitly: Use @ to reference files when asking questions (e.g., write a test for @auth.py) to avoid the model scanning the entire project unnecessarily.

• Leverage sub-agents: For large-scale tasks, let Claude spawn sub-agents to execute them. Sub-agents return refined conclusions after completing tasks, preserving context space in the main conversation.

2. Plan first, then execute

• Enable Plan mode: Before complex tasks, analyze the solution first without actually modifying files.

  • Prompt constraint: Explicitly request in your prompt to "output a detailed implementation plan first; only modify files after I confirm."

• Reduce trial-and-error costs: Ensure the logic is sound before making code changes.

3. Document core project knowledge: Write a CLAUDE.md

• Include key information: CLAUDE.md is automatically loaded at the start of each session. Include build commands, code conventions, and workflow rules.

• Maintain dynamically: Keep the content concise and readable. Only record broadly applicable global conventions, and continuously update with new rules as the project evolves.

4. Extend capabilities: MCP and Skills

• MCP: Install mature MCP servers to connect to external services.

• Skills: Write detailed skill descriptions. Claude decides whether to invoke a tool based on the definition of its purpose.

• Skills vs MCP: Skills teach Claude "how to do things" (workflow knowledge), while MCP gives Claude "the tools to do things" (external interfaces). The two are complementary, and Skills can also integrate external interfaces.

5. Automated guardrails: Hooks

• Use Hooks: Hooks are deterministic rules. They automatically run local scripts at specific lifecycle points in the Claude workflow (such as PreToolUse for pre-execution validation), ensuring critical checks or operations are executed 100% of the time.

• Configuration methods:

  1. Run /hooks for interactive configuration.

  2. Directly edit .claude/settings.json.

  3. Ask Claude to write one for you, such as: "Write a hook that runs eslint after every file edit."

6. Establish a self-verification loop

• Enforce verification: Require Claude to run relevant test cases (such as pytest or npm test) after modifying code.

• Define success criteria: "After making the changes, ensure the code compiles and run a curl command to verify the API returns a 200 status code."

• Visual feedback: When modifying the frontend, ask Claude to take browser screenshots to confirm the UI result.